Nigeria NIGSAC Sanctions MCP

Screen a person or company name against Nigeria's official sanctions register, exposed to AI agents as an MCP tool.

Status: 🌐 LIVE + keyless (no API key) Package: @braynexservices/nigeria-mcp-nigsac-sanctions v0.1.0 · depends on @braynexservices/nigeria-mcp-core Author: Samuel Orie (Founder) · Owner: Braynex Services Ltd · License: MIT © 2026 Braynex Services Ltd — MIT licensed (copyright retained; see the NOTICE).

⚠️ Screening signal, not compliance clearance. A POTENTIAL_MATCH is a prompt for human review; a CLEAR is not a legal determination. Read Scope & limitations before you rely on this for anything. Not a sole KYC/AML control.

What it does

Screens a person or company name against the official NIGSAC (Nigeria Sanctions Committee) register — the Nigerian designations of both individuals and entities, published under the Terrorism (Prevention and Prohibition) Act 2022. The list is scraped live from the NIGSAC portal (keyless), cached, and matched locally. Each screen returns CLEAR or POTENTIAL_MATCH with candidate names and 0-1 similarity scores. Public designation data — the listed names are published sanctions designations, not private PII; the query name is pass-through and never persisted.

Matching is best-effort and deliberately recall-leaning — it over-flags so a human never misses a real designation. That means a POTENTIAL_MATCH is a signal for review, not proof of a hit, and a CLEAR covers the Nigerian list only.

Tool

screen_sanctions

| Param | Required | Notes | |---|---|---| | name | ✓ | Person or company name to screen (3–256 chars) |

Returns status (CLEAR | POTENTIAL_MATCH), matches (candidates, highest similarity first — each with name, listType, similarity 0–1), listVersion (snapshot fetch-date + entry count), plus source and retrievedAt provenance.

Example — a CLEAR result (default live register):

// → screen_sanctions({ "name": "Acceptance Test Person" })
"\"Acceptance Test Person\" — CLEAR: no candidate matches on the NIGSAC Nigerian list (source: nigsac, list nigsac-2026-07-18-69). Not a legal clearance; covers the Nigerian list only."
// structuredContent:
// { query:"Acceptance Test Person", status:"CLEAR", matches:[],
//   listVersion:"nigsac-2026-07-18-69", source:"nigsac", retrievedAt }

Example — a POTENTIAL_MATCH result (shown against the offline mock stub for a reproducible hit):

// → screen_sanctions({ "name": "IBRAHIM TEST DOE" })
"\"IBRAHIM TEST DOE\" — POTENTIAL_MATCH: IBRAHIM TEST DOE (100%) — human review required (source: mock, list mock-fixtures-2026-07)"
// structuredContent:
// { query:"IBRAHIM TEST DOE", status:"POTENTIAL_MATCH",
//   matches:[{ name:"IBRAHIM TEST DOE", listType:"NIGSAC UNSC", similarity:1 }],
//   listVersion:"mock-fixtures-2026-07", source:"mock", retrievedAt }

similarity is 0–1 (1 = exact normalized match). A partial score (e.g. a query that shares most tokens with a longer listed designation) still surfaces as a candidate — by design. The percentage is a matching confidence, not a probability that the subject is sanctioned.

Configuration — SANCTIONS_PROVIDER

| Value | Behaviour | |---|---| | nigsac (default) | Live — fetches + parses the official NIGSAC register and screens against it. Keyless. This is the default; unlike other lanes, sanctions does not default to mock (a synthetic list would silently CLEAR every real name). | | mock | Offline synthetic stub — an explicit opt-in (2 fake entries, IBRAHIM TEST DOE / ACME SHELL CORP LTD) for dev/eval with zero network. Never real designations. |

No API key required.

NIGSAC_LIST_URL (optional) — override the source. Point it at either an HTML list page (parsed like the default portal) or a founder-hosted JSON array of { name, listType } (auto-detected by a leading [). Useful if the portal layout changes or you host a mirror.

Scope & limitations

Read this before relying on a result. This lane is a screening aid, not a compliance product.

• It deliberately over-flags. Matching is best-effort fuzzy: diacritic-folded (Bàbátúndé matches BABATUNDE), transliteration-tolerant (Mohammed ~ Muhammad), and containment-scored (a shorter query still scores high against a longer listed designation instead of being diluted). A false POTENTIAL_MATCH costs a human a moment; a false CLEAR misses a real designation — so the scorer leans to recall. A POTENTIAL_MATCH is a signal for human review, NOT a legal determination or compliance clearance.
• Nigerian list only. Scope is the domestic NIGSAC register. A CLEAR means "not on the Nigerian list" — not "unsanctioned everywhere". The UN consolidated list is not screened here (NIGSAC links it off-site; wiring it is a separate lane's job). Do not read a CLEAR as global.
• Fail-loud, never a false CLEAR. The live adapter refuses to screen if the register parses to an implausibly small number of entries (below a plausibility floor — the real register is ~69 rows). A changed layout, error, or challenge page raises an actionable error instead of screening against a truncated list and clearing real names.
• Freshness is visible. The list is cached with a TTL (re-fetched at most every ~6h; it changes rarely) and fetched with a 15s timeout + backoff-retry on transient failures. Every result carries a listVersion (fetch date + entry count, e.g. nigsac-2026-07-18-69) so callers can see exactly how fresh the snapshot is.
• NOT compliance-grade. Do not use this as a sole KYC/AML control. It is one screening input among many — pair any hit with authoritative verification and human judgement.

Run it

npm install && npm run build      # from repo root

npm run smoke      -w @braynexservices/nigeria-mcp-nigsac-sanctions   # offline (mock stub)
npm run smoke:live -w @braynexservices/nigeria-mcp-nigsac-sanctions   # live (real NIGSAC register)

npm run start:stdio -w @braynexservices/nigeria-mcp-nigsac-sanctions  # serve over stdio (live by default)

MCP Inspector (interactive GUI):

npx @modelcontextprotocol/inspector node lanes/nigsac-sanctions/dist/stdio.js

Claude Desktop / Code — add to your MCP config (needs Node.js ≥ 20):

{
  "mcpServers": {
    "nigeria-nigsac-sanctions": {
      "command": "npx",
      "args": ["-y", "@braynexservices/nigeria-mcp-nigsac-sanctions"]
    }
  }
}

The default is the live register — no env block is needed. To run fully offline against the synthetic stub instead, add "env": { "SANCTIONS_PROVIDER": "mock" }. Running from a clone: "command": "node", "args": ["/path/to/nigeria-mcp/lanes/nigsac-sanctions/dist/stdio.js"]. Then ask, e.g., "Screen 'Ada Umeh Ltd' against the Nigerian sanctions list."

Verification

npm run verify (repo root) → build + all lane smokes + eval. This lane: an offline end-to-end smoke (exact / fuzzy-containment / CLEAR paths vs the mock stub) plus smoke:live against the real NIGSAC register (verified live 2026-07-18: HTTP 200, two tables — individuals + entities — 69 entries; parse quality confirmed against real rows). The live adapter's plausibility floor and cache/retry behaviour are exercised in review.